Applied Machine Learning
Autoencoder can be a powerful data preprocessing tool for art investigation where annotation data is limited. By extending partial convolution, we constructed a fully partial convolutional autoencoder (FP-CAE) and adapted it to multimodal data. We introduced a SSIM-based loss function to train the autoencoder. Experimental results on the Ghent Altarpiece show that our method significantly suppresses edge artifacts and improves the overall reconstruction performance.
Xianghui Xie is currently enrolling in a 2+2 program in KU Leuven, majoring in Electronics Engineering. This is the second year he studies in Belgium, also the final year as an undergraduate student. Prior to that, he spent two years in Beijing Jiao Tong University, where he majored in Telecommunication Engineering.